Implementasi Dan Analisis YOLOv4 Pada Raspberry Pi untuk Estimasi Jarak dan Deteksi Objek Dalam Kendaraan Otonom

Montella, Michael (2023) Implementasi Dan Analisis YOLOv4 Pada Raspberry Pi untuk Estimasi Jarak dan Deteksi Objek Dalam Kendaraan Otonom. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07111940000195-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (2MB) | Request a copy

Abstract

Studi ini merinci implementasi mounting kamera pada mobil otonom menggunakan Raspberry Pi 3B sebagai perangkat pengambilan data. Data dikumpulkan dan diproses secara offline pada laptop. Kalibrasi distorsi kamera dilakukan menggunakan pola papan catur 9×6 yang dicetak pada kertas A4, dan matriks distorsi diterapkan untuk koreksi gambar. Mounting kamera, terbuat dari besi dan almini, dilengkapi dengan suction cup untuk pemasangan pada kap mobil. Raspberry Pi ditenagai oleh baterai Lipo Gens Ace 5000mAH dan dikontrol secara remote melalui koneksi SSH yang dihubungkan dengan router wifi. Algoritma YOLOv4 ditraining dengan 2 kelas, yaitu car dan motorbike, dengan F1 score sebesar 0.7476. Estimasi jarak dilakukan dengan interval 3 meter hingga 30 meter, menggunakan pendekatan pinhole camera, menghasilkan RMSE 1.69 m. Kecepatan relatif dihitung dengan data gambar yang diambil 2 fps selama 20 detik. Kecepatan rata-rata mobil yang diamati dan mobil dengan mounting didapatkan dari GPS logger dan didapatkan nilai kecepatan rata-rata 5.1 km/jam pada mobil yang diamati dam 5 km/jam pada mobil dengan mounting. Jarak aman untuk kecepatan mobil yang mengamati adalah 1.875 m. Estimasi kecepatan relatif yang didapat adalah 0.0666 km/ jam dan mobil yang digunakan untuk mengamati berhenti pada jarak 3.77 m, sehingga masih tergolong jarak aman.
===================================================================================================================================
This study details the implementation of a camera mount on an autonomous car using a Raspberry Pi 3B as the data collection device. Data was collected and processed offline on a laptop. Camera distortion calibration was performed using a 9x6 chessboard pattern printed on A4 paper, and a distortion matrix was applied for image correction. The camera mount, made of iron and aluminum, was equipped with a suction cup for attachment to the car hood. The Raspberry Pi was powered by a Gens Ace 5000mAH Lipo battery and was remotely controlled via an SSH connection connected to a wifi router. The YOLOv4 algorithm was trained with two classes, car and motorbike, with an F1 score of 0.7476. Distance estimation was conducted with intervals of 3 meters up to 30 meters, using the pinhole camera approach, resulting in an RMSE of 1.69 m. Relative speed was calculated using image data taken at 2 fps for 20 seconds. The average speed of the observed car and the car with the mount was obtained from a GPS logger, yielding average speeds of 5.1 km/h for the observed car and 5 km/h for the car with the mount. The safe distance for the observing car's speed was 1.875 m. The estimated relative speed obtained was 0.0666 km/h, and the observing car stopped at a distance of 3.77 m, thus still being within the safe distance range.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kendaraan Otonom, YOLOv4, Raspberry Pi, Pinhole Camera, Estimasi Jarak, Estimasi Kecepatan rata-rata, Deteksi Objek, YOLOv4, Raspberry Pi, Pinhole Camera, Object Detection, Distance Estimation, Autonomous Vehicles, Average Speed Estimation.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL152.8 Vehicles, Remotely piloted. Autonomous vehicles.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Michael Montella
Date Deposited:	25 Jul 2023 04:16
Last Modified:	25 Jul 2023 04:16
URI:	http://repository.its.ac.id/id/eprint/99264

Actions (login required)

View Item